The University of Edinburgh -
Division of Informatics
Forrest Hill & 80 South Bridge

MSc Thesis #93102

Title:Computer Generation of Linguistically Definable Riddles
Date: 1993
Abstract:In this dissertation, we discuss a model of simple question-answer punning riddles, developed over the course of the project, which we have implemented on computer. the resulting program, JAPE (Joke Analysis and Production Engine), generates jokes from humour-independent lexical entries. After reviewing the limited literature in the area, we make a rough taxonomy of riddle types, based on the levels of ambiguity they have and the mechanisms they use to exploit that ambiguity. Having chosen a well-defined subgroup of riddles - question-answer punning riddles - we describe a model for that type of joke. The model uses two main types of structure: a schemata, which determine the relationships between key words in a joke, and templates, which produce the surface form of the joke. We then describe JAPE's design, and evaluate its performance. Although JAPE succeeds in generating pieces of text that are recognizably jokes, they are not very good jokes. We suggest further improvements and extensions; in particular, the development of post-production heuristics for ordering the jokes to quality, and the extension of the model to cover other types of text-based humour.

[Search These Pages] [DAI Home Page] [Comment]